A distilled version of Whisper with 2 decoder layers, optimized for French speech-to-text.

Compared to v0.1, this version extends the training to 30-second audio segments to maintain long-form transcription abilities. The training process used a "patient" teacher during distillation - meaning longer training times and more aggressive data augmentation - which improved overall performance.

The model uses openai/whisper-large-v3 as the teacher model while keeping the encoder architecture unchanged. This makes it suitable as a draft model for speculative decoding, potentially getting 2x inference speed while maintaining identical outputs by only adding 2 extra decoder layers and running the encoder just once. It can also serve as a standalone model to trade some accuracy for better efficiency, running 5.8x faster while using only 49% of the parameters. This paper also suggests that the distilled model may actually produce fewer hallucinations than the full model during long-form transcription.

The model has been converted into multiple formats to ensure broad compatibility across libraries including transformers, openai-whisper, faster-whisper, whisper.cpp, candle, mlx.

The model was evaluated on both short and long-form transcriptions, using in-distribution (ID) and out-of-distribution (OOD) datasets to assess accuracy, generalizability, and robustness.

Note that Word Error Rate (WER) results shown here are post-normalization, which includes converting text to lowercase and removing symbols and punctuation.

All evaluation results on the public datasets can be found here.

Italic indicates in-distribution (ID) evaluation, where test sets correspond to data distributions seen during training, typically yielding higher performance than out-of-distribution (OOD) evaluation. Italic and strikethrough denotes potential test set contamination - for example, when training and evaluation use different versions of Common Voice, raising the possibility of overlapping data.

Due to the limited availability of out-of-distribution (OOD) and long-form French test sets, evaluation was also performed using internal test sets from Zaion Lab - consisting of human-annotated call center conversations with significant background noise and domain-specific terminology.

Long-form transcription evaluation used the 🤗 Hugging Face pipeline with both chunked (chunk_length_s=30) and original sequential decoding methods.

The model can be easily used with the 🤗 Hugging Face pipeline class for audio transcription. For long-form transcription (over 30 seconds), it will perform sequential decoding as described in OpenAI's paper. If you need faster inference, you can use the chunk_length_s argument for chunked parallel decoding, which provides 9x faster inference speed but may slightly compromise performance compared to OpenAI's sequential algorithm.

You can also use the 🤗 Hugging Face low-level APIs for transcription, offering greater control over the process, as demonstrated below:

Speculative decoding can be achieved using a draft model, essentially a distilled version of Whisper. This approach guarantees identical outputs to using the main Whisper model alone, offers a 2x faster inference speed, and incurs only a slight increase in memory overhead.

Since the distilled Whisper has the same encoder as the original, only its decoder need to be loaded, and encoder outputs are shared between the main and draft models during inference.

Using speculative decoding with the Hugging Face pipeline is simple - just specify the assistant_model within the generation configurations.

You can also employ the sequential long-form decoding algorithm with a sliding window and temperature fallback, as outlined by OpenAI in their original paper.

First, install the openai-whisper package:

Then, download the converted model:

Now, you can transcirbe audio files by following the usage instructions provided in the repository:

Faster Whisper is a reimplementation of OpenAI's Whisper models and the sequential long-form decoding algorithm in the CTranslate2 format.

Compared to openai-whisper, it offers up to 4x faster inference speed, while consuming less memory. Additionally, the model can be quantized into int8, further enhancing its efficiency on both CPU and GPU.

First, install the faster-whisper package:

Then, download the model converted to the CTranslate2 format:

Now, you can transcirbe audio files by following the usage instructions provided in the repository:

Whisper.cpp is a reimplementation of OpenAI's Whisper models, crafted in plain C/C++ without any dependencies. It offers compatibility with various backends and platforms.

Additionally, the model can be quantized to either 4-bit or 5-bit integers, further enhancing its efficiency.

First, clone and build the whisper.cpp repository:

Next, download the converted ggml weights from the Hugging Face Hub:

Now, you can transcribe an audio file using the following command:

Candle-whisper is a reimplementation of OpenAI's Whisper models in the candle format - a lightweight ML framework built in Rust.

First, clone the candle repository:

Transcribe an audio file using the following command:

In order to use CUDA add --features cuda to the example command line:

MLX-Whisper is a reimplementation of OpenAI's Whisper models in the MLX format - a ML framework on Apple silicon. It supports features like lazy computation, unified memory management, etc.

First, clone the MLX Examples repository:

Next, install the dependencies:

Download the pytorch checkpoint in the original OpenAI format and convert it into MLX format (We haven't included the converted version here since the repository is already heavy and the conversion is very fast):

Now, you can transcribe audio with:

We built a French speech recognition dataset of over 22,000 hours of annotated and semi-annotated speech. After decoding this dataset through Whisper-Large-V3 and filtering out segments with WER above 20%, we retained approximately 10,000 hours of high-quality audio.

Most data were first concatenated into 30-second segments, primarily preserving the same speaker, then inferred together. 50% of segments were trained with timestamps to ensure good timestamp prediction, and only 20% of segments were trained with previous context since we don't expect the 2-layer decoder to excel at this task.

This model was trained for a fairly long schedule of 160 epochs using aggressive data augmentation, with eval WER continuing to decrease. Some hyperparameter choices were made to favor long-form over short-form transcription. For further details, please refer to the Distil-Whisper repository.

All model training was conducted on the Jean-Zay supercomputer at GENCI. Special thanks to the IDRIS team for their excellent support throughout this project.